fs/ocfs2/cluster/tcp_internal.h

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * Copyright (C) 2005 Oracle.  All rights reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2 of the License, or (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public
  17  * License along with this program; if not, write to the
  18  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  19  * Boston, MA 021110-1307, USA.
  20  */
  21
  22 #ifndef O2CLUSTER_TCP_INTERNAL_H
  23 #define O2CLUSTER_TCP_INTERNAL_H
  24
  25 #define O2NET_MSG_MAGIC           ((u16)0xfa55)
  26 #define O2NET_MSG_STATUS_MAGIC    ((u16)0xfa56)
  27 #define O2NET_MSG_KEEP_REQ_MAGIC  ((u16)0xfa57)
  28 #define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
  29
  30 /* same as hb delay, we're waiting for another node to recognize our hb */
  31 #define O2NET_RECONNECT_DELAY_MS        O2HB_REGION_TIMEOUT_MS
  32
  33 /* we're delaying our quorum decision so that heartbeat will have timed
  34  * out truly dead nodes by the time we come around to making decisions
  35  * on their number */
  36 #define O2NET_QUORUM_DELAY_MS   ((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
  37
  38 #define O2NET_KEEPALIVE_DELAY_SECS      5
  39 #define O2NET_IDLE_TIMEOUT_SECS         10
  40
  41 /*
  42  * This version number represents quite a lot, unfortunately.  It not
  43  * only represents the raw network message protocol on the wire but also
  44  * locking semantics of the file system using the protocol.  It should
  45  * be somewhere else, I'm sure, but right now it isn't.
  46  *
  47  * New in version 2:
  48  *      - full 64 bit i_size in the metadata lock lvbs
  49  *      - introduction of "rw" lock and pushing meta/data locking down
  50  */
  51 #define O2NET_PROTOCOL_VERSION 2ULL
  52 struct o2net_handshake {
  53         __be64  protocol_version;
  54         __be64  connector_id;
  55 };
  56
  57 struct o2net_node {
  58         /* this is never called from int/bh */
  59         spinlock_t                      nn_lock;
  60
  61         /* set the moment an sc is allocated and a connect is started */
  62         struct o2net_sock_container     *nn_sc;
  63         /* _valid is only set after the handshake passes and tx can happen */
  64         unsigned                        nn_sc_valid:1;
  65         /* if this is set tx just returns it */
  66         int                             nn_persistent_error;
  67
  68         /* threads waiting for an sc to arrive wait on the wq for generation
  69          * to increase.  it is increased when a connecting socket succeeds
  70          * or fails or when an accepted socket is attached. */
  71         wait_queue_head_t               nn_sc_wq;
  72
  73         struct idr                      nn_status_idr;
  74         struct list_head                nn_status_list;
  75
  76         /* connects are attempted from when heartbeat comes up until either hb
  77          * goes down, the node is unconfigured, no connect attempts succeed
  78          * before O2NET_CONN_IDLE_DELAY, or a connect succeeds.  connect_work
  79          * is queued from set_nn_state both from hb up and from itself if a
  80          * connect attempt fails and so can be self-arming.  shutdown is
  81          * careful to first mark the nn such that no connects will be attempted
  82          * before canceling delayed connect work and flushing the queue. */
  83         struct work_struct              nn_connect_work;
  84         unsigned long                   nn_last_connect_attempt;
  85
  86         /* this is queued as nodes come up and is canceled when a connection is
  87          * established.  this expiring gives up on the node and errors out
  88          * transmits */
  89         struct work_struct              nn_connect_expired;
  90
  91         /* after we give up on a socket we wait a while before deciding
  92          * that it is still heartbeating and that we should do some
  93          * quorum work */
  94         struct work_struct              nn_still_up;
  95 };
  96
  97 struct o2net_sock_container {
  98         struct kref             sc_kref;
  99         /* the next two are vaild for the life time of the sc */
 100         struct socket           *sc_sock;
 101         struct o2nm_node        *sc_node;
 102
 103         /* all of these sc work structs hold refs on the sc while they are
 104          * queued.  they should not be able to ref a freed sc.  the teardown
 105          * race is with o2net_wq destruction in o2net_stop_listening() */
 106
 107         /* rx and connect work are generated from socket callbacks.  sc
 108          * shutdown removes the callbacks and then flushes the work queue */
 109         struct work_struct      sc_rx_work;
 110         struct work_struct      sc_connect_work;
 111         /* shutdown work is triggered in two ways.  the simple way is
 112          * for a code path calls ensure_shutdown which gets a lock, removes
 113          * the sc from the nn, and queues the work.  in this case the
 114          * work is single-shot.  the work is also queued from a sock
 115          * callback, though, and in this case the work will find the sc
 116          * still on the nn and will call ensure_shutdown itself.. this
 117          * ends up triggering the shutdown work again, though nothing
 118          * will be done in that second iteration.  so work queue teardown
 119          * has to be careful to remove the sc from the nn before waiting
 120          * on the work queue so that the shutdown work doesn't remove the
 121          * sc and rearm itself.
 122          */
 123         struct work_struct      sc_shutdown_work;
 124
 125         struct timer_list       sc_idle_timeout;
 126         struct work_struct      sc_keepalive_work;
 127
 128         unsigned                sc_handshake_ok:1;
 129
 130         struct page             *sc_page;
 131         size_t                  sc_page_off;
 132
 133         /* original handlers for the sockets */
 134         void                    (*sc_state_change)(struct sock *sk);
 135         void                    (*sc_data_ready)(struct sock *sk, int bytes);
 136
 137         struct timeval          sc_tv_timer;
 138         struct timeval          sc_tv_data_ready;
 139         struct timeval          sc_tv_advance_start;
 140         struct timeval          sc_tv_advance_stop;
 141         struct timeval          sc_tv_func_start;
 142         struct timeval          sc_tv_func_stop;
 143         u32                     sc_msg_key;
 144         u16                     sc_msg_type;
 145 };
 146
 147 struct o2net_msg_handler {
 148         struct rb_node          nh_node;
 149         u32                     nh_max_len;
 150         u32                     nh_msg_type;
 151         u32                     nh_key;
 152         o2net_msg_handler_func  *nh_func;
 153         o2net_msg_handler_func  *nh_func_data;
 154         struct kref             nh_kref;
 155         struct list_head        nh_unregister_item;
 156 };
 157
 158 enum o2net_system_error {
 159         O2NET_ERR_NONE = 0,
 160         O2NET_ERR_NO_HNDLR,
 161         O2NET_ERR_OVERFLOW,
 162         O2NET_ERR_DIED,
 163         O2NET_ERR_MAX
 164 };
 165
 166 struct o2net_status_wait {
 167         enum o2net_system_error ns_sys_status;
 168         s32                     ns_status;
 169         int                     ns_id;
 170         wait_queue_head_t       ns_wq;
 171         struct list_head        ns_node_item;
 172 };
 173
 174 #endif /* O2CLUSTER_TCP_INTERNAL_H */